Bioinformatics Advance Access originally published online on September 23, 2007
Bioinformatics 2007 23(20):2725-2732; doi:10.1093/bioinformatics/btm425
Fun&Co: identification of key functional differences in transcriptomes
1DAMA - Data Mining for Analysis of DNA microarrays and GebbaLab - Department of Morphology and Embryology, Via Fossato di Mortara 64/b - 44100 Ferrara and 2ENDIF, Department of Engineering, Via Saragat 1 - 44100 Ferrara, Italy
*To whom correspondence should be addressed.
 |
ABSTRACT |
|---|
 TOP ABSTRACT 1 INTRODUCTION2 APPROACH3 METHODS4 RESULTS AND DISCUSSION5 CONCLUSIONACKNOWLEDGEMENTSREFERENCES |
|---|
Motivation: Microarray and other genome-wide technologies allow a global view of gene expression that can be used in several ways and whose potential has not been yet fully discovered. Functional insight into expression profiles is routinely obtained by using gene ontology terms associated to the cellular genes. In this article, we deal with functional data mining from expression profiles, proposing a novel approach that studies the correlations between genes and their relations to Gene Ontology (GO). We implemented this approach in a public web-based application named Fun&Co. By using Fun&Co, the user dissects in a pair-wise manner gene expression patterns and links correlated pairs to gene ontology terms. The proof of principle for our study was accomplished by dissecting molecular pathways in muscles. In particular, we identified specific cellular pathways by comparing the three different types of muscle in a pairwise fashion. In fact, we were interested in the specific molecular mechanisms regulating the cardiovascular system (cardiomyocytes and smooth muscle cells).
Results: We applied here Fun&Co to the molecular study of cardiovascular system and the identification of the specific molecular pathways in heart, skeletal and smooth muscles (using 317 microarrays) and to reveal functional differences between the three different kinds of muscle cells.
Availability: Application is online at http://tommy.unife.it.
Contact: s.volinia{at}unife.it
Supplementary information: Supplementary data are available at Bioinformatics online.
|
1 INTRODUCTION |
|---|
|
TOPABSTRACT1 INTRODUCTION2 APPROACH3 METHODS4 RESULTS AND DISCUSSION5 CONCLUSIONACKNOWLEDGEMENTSREFERENCES |
|---|
For a thorough understanding of transcriptome-wide experiments, it is essential to extract essential information from expression datasets in the most efficient and straightforward way.
DNA microarray and other high-throughput technical platforms generate genome-wide gene expression data (Eisen et al., 1999). Each gene product may be involved in one or more functions. These functions are accurately described in the Gene Ontology database. The Gene Ontology (GO) consortium1 produces three independent ontologies for gene products. The three ontologies form the basis to describe the molecular function, the biological process and the cellular component of gene products.
In this article, we devised a novel method to use GO terms and differentiate biological samples, such as tissues or cell lines. We implemented the method and showed here its application to elucidate the intricate web of signals which regulate human muscle function.
In particular, we wished to understand which key pathways are peculiar to a very important human system: the cardiovascular system. In order to identify the relevant specific functional modules, we performed a comprehensive comparison of heart and smooth muscle cells with skeletal muscle. Skeletal muscle forms the bulk of the body's; muscle tissue. Smooth muscle, a second type of muscle, which lines most of the hollow organs of the body, is not under voluntary control, but is regulated by the autonomic nervous system. Cardiac muscle is striated like skeletal muscle but, like the smooth muscle, is involuntarily controlled.
The method described in this article investigates the correlations between genes and their GO terms. Given a general term of the GO tree, our algorithm selects the genes related to this term, or the more specific ones. Considering the selected genes, Fun&Co computes the correlation between each gene-pair expression and identifies the pairs with a correlation level higher than a threshold. P-values are corrected by means of false detection rate (FDR). Then the application compares two or more datasets, searching for the GO terms with different behavior among samples.
We used datasets produced on Affymetrix U133 platforms (U133a,b or U133plus) and available from the GEO public data repository.2 In particular, we used for heart 107 samples from GSE1145 [NCBI GEO] and 70 samples from GSE2240 [NCBI GEO] and for skeletal muscle 79 samples from GSE3307 [NCBI GEO] and 35 from GSE4667 [NCBI GEO] . For the smooth muscle dataset, we merged the data from GSE1595 [NCBI GEO] , GSE2883 [NCBI GEO] and GSE3356 [NCBI GEO] in a single table, obtaining a 26 samples dataset.
The goal of our model study was that of identifying specific cellular pathways by comparing the three types of muscle in a pairwise fashion. In particular, we were interested in the specific molecular mechanisms regulating the cardiovascular system (cardiomyocytes and smooth muscle cells).
1.1 Related work
In literature, a number of methods have been described for GO enrichment analysis. For example, some GO-oriented analyses are GOAL, EASE, Onto-Express, GSEA, FatiScan, GlobalTest and ErmineJ. These software packages are designed to help biologists with the interpretation of genome-scale data.
GOAL (Gene Ontology Automated Lexicon) (Volinia et al., 2004) is a web resource for automated and streamlined functional analysis of expression profiles. It aims to detect those GO terms which are significantly regulated. It automatically generates and evaluates scoring of GO terms from the results of an expression-profiling experiment. Permutation analysis is performed to define P-values and false detection rates within each dataset.
EASE (Hosack et al., 2003) performs theme discovery with any list of genes. Theme discovery is defined as the identification of terms or phrases that describe a statistically significant number of genes in the list with respect to the number of genes described by the term or phrase in the population of genes from which the list is derived.
Onto-Express (Draghici et al., 2003) is a tool able to translate lists of differentially regulated genes into functional profiles. It is very similar to EASE, and can manage several functional categories, using GO terms. OE provides implementation of different statistics to calculate a P-value for each functional category (binomial test, 
2 test and Fisher's; exact test).
Gene Set Enrichment Analysis (GSEA) (Mootha et al., 2003) is a tool that analyzes a dataset with two classes of samples. The genes are ranked based on the correlation between their expression and the class distinction. Then, GSEA determines whether the genes related to a particular function are randomly distributed throughout the ranked list, or primarily found at the top or bottom.
FatiScan (Al-Shahrour et al., 2004, 2005) has an approach very similar to GSEA, but implements a more sensible statistic and can reveal different types of asymmetries in the distributed groups of genes.
GlobalTest (Goeman and van Houwelingen, 2004) is an R-package for bioconductor, it is based on a prediction model for predicting a response variable from the gene expression measurements of a set of genes; genes can be grouped together into genesets, e.g. based in function (KEGG, GO) or location (chromosome, cytoband). GlobalTest compares group of samples (e.g. CLL versus ALL) by using genes expression profiles; to test a specific pathway, the user can provide it as argument to globaltest. GlobalTest by using expression profile of a specific pathway (e.g. specific GO term) can provide a measure of difference between different group of samples by computing a P-value.
ErmineJ (Lee et al., 2005) is a software tool that analyzes functionally relevant sets of genes through microarray gene expression data. ErmineJ provides different classical type of GO-based analysis: hypergeometric distribution or its binomial approximation and GSR (Gene Set Resampling) that computes gene-by-gene non-thresholded scores determined by random sampling (Pavlidis et al., 2002) of the data. In addition, ErmineJ provides a ranking-based analysis and an analysis based on the correlation of gene expression profiles. ErmineJ does not evidence differences between different classes of samples, while Fun&Co method is classification oriented, and it aims to highlight these differences.
Most of these applications analyze only the statistical difference between two lists of differentially expressed genes, derived from the datasets. The approach we used for Fun&Co and we describe here is in principle fairly different, as: (i) it is based on gene pairs rather than gene singlets, and (ii) it measures the correlation between expression levels, rather then differential expression.
|
2 APPROACH |
|---|
|
TOPABSTRACT1 INTRODUCTION2 APPROACH3 METHODS4 RESULTS AND DISCUSSION5 CONCLUSIONACKNOWLEDGEMENTSREFERENCES |
|---|
Fun&Co is an application for dissecting and comparing expression profiles at a functional level. We developed and applied it to the identification of molecular differences in the three skeletal, cardiac and smooth muscles. This study represents both a test model and a very important scientific and medical problem. Cardiovascular diseases represent one of the most common and serious health issues. Thus, a fine understanding of the molecular mechanisms underlying heart physiology is of great importance. We used here, the wealth of data generated by a number of laboratories on muscle transcriptomes to identify key functions and processes in human heart, and smooth cells when compared to each other and to the skeletal muscle. The rationale is that the specific functions and pathways will be of relevant medical importance.
By linking the patterns of gene-pairs expression to the respective gene function (as provided by the Gene Ontology database), we can extract information to better understand genome-wide expression profiles and to help scientists in the subsequent design of focused experiments. As a proof-of-principles, we identified the GO terms which distinguish different tissues.
In details, the functional correlations comparison aims to highlight changes in gene expression correlations, in order to identify relations involved in tissue differentiation. Merging these results with the GO annotations, we can immediately select functionally relevant biological entities associated to different tissues.
We used the Spearman's; correlation coefficient (Rosner, 1995) (described in ‘Methods’ section) to evaluate the correlation between the mRNA levels of all possible gene pairs. Finally, we linked these results with the GO terms and selected the significant functional differences.
The approach, shown in Figure 1, consists of four steps: (i) gene selection; (ii) correlation computation; (iii) dataset comparison and (iv) result synthesis. These steps are described in details in the ‘Methods’ section.
|
|
3 METHODS |
|---|
|
TOPABSTRACT1 INTRODUCTION2 APPROACH3 METHODS4 RESULTS AND DISCUSSION5 CONCLUSIONACKNOWLEDGEMENTSREFERENCES |
|---|
This section describes in detail all the steps used for identifying functional differences between different tissues by analyzing a pair of dataset groups, e.g. muscle and heart.
3.1 Dataset selection
In this first step, the users have to choose between the available GEO datasets the ones to use in the Fun&Co analysis. It is also possible to upload users' owned datasets.
These datasets are partitioned in a pairwise manner (FA1,..., FAn) and FB1, ..., FBm) depending on the kind of analysis (e.g. different tissues, cancer and normal samples, treated and normal samples).
3.2 Gene selection
Given the two dataset groups, the users have to specify a GO term (named GOmain). Fun&Co uses GOmain to filter the probes contained in the datasets, keeping only the probes annotated with such GO term or a more specific one (probe annotations are taken from the Affymetrix support library).
This is necessary for two reasons: (i) to focus the analysis only on probes related to GOmain and its descendants; (ii) to reduce the computation workload of the next steps. The filtered datasets are named A1, ..., An and B1, ..., Bm.
3.3 Correlation study
After the gene selection described in the previous section, Fun&Co studies any possible pair of probes in the dataset, valuing the correlation coefficient of each pair (for measuring the degree of relationship between two variables, i.e. two probes).
Due to the nature of the Affymetrix microarrays, it is preferable to use a non-parametric correlation coefficient. For this reason, our tool uses Spearman's; correlation coefficient instead of Pearson's; (as described in Section 3.3).
In principle, Spearman's; correlation coefficient (usually denoted by 
) is simply a special case of the Pearson product-moment coefficient, in which the data are converted into ranks before calculating the coefficient (when converting into ranks, the smallest value becomes a rank of 1, the second becomes 2, etc.). The raw scores are converted into ranks, and the differences d between the ranks of each observation on the two variables are calculated. The coefficient can also be computed using this formula;
|
| (1) |
3.3.1 Significance level of correlation
For each probe pair, the significance level of the Spearman's; correlation was evaluated computing the P-value. The evaluation starts from the assumption that the correlation values were distributed using Student's; t cumulative distribution, with a number of degrees of freedom corresponding to the number of samples in the microarray experiment minus 2. First, a t value was computed by using the following equation:
|
| (2) |
is the correlation coefficient and n is the number of samples. Then, looking for t in the Student's; t cumulative distribution table for n – 2 degrees of freedom, the P-value of each probe pair was found.
3.3.2 False detection rate
Given the high number of correlation tests performed, P-values can be corrected for multiple testing by using the FDR (Benjamini and Hochberg, 1995). FDR controls a different probability than the one controlled with the better known P-value. In fact, P-values control the number of false positive out of the number of truly null tests, while FDR controls the number of false positive over the number of significant tests.
Several ways of estimating this number have been proposed, we adopted the solution devised by Tom Nichols,3 which rescales the P-value obtained on a single test multiplying it by a combination of indexes related to the total number of tests performed: 
, where pi represents the i-th of the total K single P-values. Correction was performed on the complete list of P-values, i.e. one value for each computed correlation.
3.3.3 Correlation filtering
Users can decide when a probe pair correlation is significant or not by choosing a p-threshold (default is 0.05): if the P-value is less than the p-threshold, then the probes are considered correlated.
3.4 Group comparison based on correlation
The aim of the group comparison was to identify the GO terms that showed a significant difference between the two groups. This was performed in three steps. In the first step, for each dataset, Fun&Co studied the correlations as described in Section 3.3. In the second step, for each dataset, the system counted the number of correlated pairs of probesets associated to the same GO term. In the last step, for each GO term, the system compared the number of correlated probe pairs detected in the two groups under study.
3.4.1 Correlation counting
As described in Section 3.3, for each dataset a correlation study was performed, leading to the identification of a set of well-correlated probe pairs. Given this set of pairs, for each GO term considered in the analysis (GOmain or one of its descendants), the system counted the number of pairs associated to this term. A pair was associated to a GO term if both probes were annotated with this GO term (probe annotations were taken from the Affymetrix support library4). In this step, we did not count pairs between probesets belonging to the same gene, because these would be trivial relations.
3.4.2 Comparison between two datasets
Given two gene expression datasets, named A and B, for each GO term under analysis, Fun&Co computed the number of correlated probe pairs found in each dataset [named N_couples(A)) and N_couples(B)]. At this point, the system discarded all the GO terms that had too few probe pairs in both datasets: a GO term was discarded if its pair count is below a user-defined threshold, named Fitnessth, in both datasets. This filtering was performed in order to avoid results that were not supported by a minimum number of found correlations. For each GO term not discarded by Fitnessth, Fun&Co identified if this term in the A dataset was over-correlated (under-correlated) with respect to the B dataset. In order to identify if a GO term was over-correlated or under-correlated, Fun&Co computed the logratio measure: 
. Then it normalized this value, subtracting to it the mean, computed among all the GO terms. By introducing a threshold (t), the GO terms over-correlated (under-correlated) in the Da samples with respect to the Db samples were the ones with 
(
).
3.4.3 Dataset group comparison
As described in Section 3.1, Fun&Co may search for functional differences between two groups of datasets (A1, ..., An and B1, ..., Bm) rather than simply considering two datasets. This feature requires an extension of the analysis approach described in Section 3.4.2.
Fun&Co performed the extended analysis, combining the results of n x m comparisons (each dataset Ai of the first group compared with each dataset Bj from the second group). The results of these comparisons were collected in a table named ‘continuous comparison’, that had n x m columns (where n is the number of datasets in the first group and m is the number of datasets in the second group).
As described before, Fun&Co considered only the GO terms with at least Fitnessth correlated probe pairs in at least one dataset. After computing the normalized logratio, it identified the over-correlated GO terms as described in Section 3.4.2.
In order to provide synthetic results, Fun&Co built a ‘consensus list’ table, that includes all the GO terms over-correlated for at least 50% of comparisons. For each GO term, it provided also the group in which this term was over-correlated and the mean of its normalized logratio in all performed comparisons.
3.5 Implementation
The tool was structured as a web server which manages Fun&Co computations (also named job) requested by the users. The web interface was developed using JSP and the computation core was developed under JAVA. The user interface allows user registration, job submission and results retrieval. Users accounts and job requests were stored in a database (MySql).
First of all, the user separately uploads datasets of the two groups and assigns a name to each group. Then, the user sets the False Detection Rate check box (if he/she wants to activate FDR adjustment of the P-value), and the p-threshold value. Finally, the user has to select a target GO from the Gene Ontology database (BUILD GO_20060115). In order to ease the GO selection, a search facility is offered in the web page. During a job execution, the progress is visualized through a status window.
The computation core was implemented as a server daemon that manages the scheduled user requests, retrieving them from a queue. Since running these computations can require several minutes, an e-mail service was provided that sends to the user a notification when the process is finished. Attached to this e-mail notification, the user finds the following output:
- ‘Single datasets results’ that contain the number of correlated probe pairs found in each dataset for all the GO terms considered (counted as described in Section 3.4.1);
- ‘Continuous comparison’ table (plain text file): this table contains for each GO term the normalized logratio value for each dataset comparison (as described in Section 3.4.2, if we have n datasets in the A group and m in the B group, the number of comparison values is n x m);
- ‘Consensus list’, that includes all the unbalanced GO terms (identified as described in Section 3.4.3);
- for each comparison, a scatter plot image (example shown in Fig. 2), which presents on the x-axis the number of couples found in the dataset of the A group, on the y-axis the number of couples found in the dataset of the B group, and a point for each GO term considered (both axes are in log scale).
|
|
4 RESULTS AND DISCUSSION |
|---|
|
TOPABSTRACT1 INTRODUCTION2 APPROACH3 METHODS4 RESULTS AND DISCUSSION5 CONCLUSIONACKNOWLEDGEMENTSREFERENCES |
|---|
Our study was aimed to identify specific cellular pathways by comparing the three different muscles in a pairwise fashion. So we performed three pairwise comparisons between the datasets corresponding to the three tissues. Some GO terms related to general functions, processes and components were chosen, to avoid exploring all the possible and not relevant GO terms for the muscular tissues (see Table 1). In particular, we investigated the response to stimuli and extracellular environments.
|
For each of these GO terms, we performed all three possible pairwise comparisons (heart versus skeletal, heart versus smooth and skeletal versus smooth) applying Fun&Co to the datasets presented in Section 1. We used the default P-value (0.05) with the FDR adjustment and set Fitnessth = 5 and LogRatioth = 1. We obtained three consensus lists (one for each comparison). The number of significant terms in the consensus lists is shown in Figure 3.
|
In order to assess the significance of the terms included in the consensus lists, we performed a bootstrap test. We randomly re-assigned the association table between the probeset ids and the GO terms, and generate 1000 bootstrapped annotation tables. We computed the consensus lists for each association table and use these results to define the P-value of each term found in the consensus lists. The P-value is computed as the ratio between the number of times in which a term is included in the consensus list and 1000 (the total number of bootstrapped tables). This kind of test is very similar to the comparative p computed by GlobalTest.
Briefly, comparative p is a useful diagnostic to see whether a gene set found by GlobalTest is exceptionally significant. It gives a fraction of random gene sets of the same size as the ‘real’ input GO term which has a larger standardized test statistic than the ‘real’ GO term itself. Pathways of the same size with a larger standardized test statistic will almost invariably also have a lower P-value.5
We were able to perform only a partial comparison with GlobalTest, because our tool is aimed to find the GO terms most involved in tissue differentiation, while GlobalTest takes the GO term to analyze as an input. For some GO terms well supported by literature we compared the results achieved by Fun&Co with the GlobalTest ones. In particular, we identified if they both identify a GO term as significant (term included in the consensus list for Fun&Co and associated with a P-value lesser than 0.001 for Globaltest) and compared the suitability of the methods in evaluating the GO under analysis (permutation test in Fun&Co and comparative p in GlobalTest).
In the results tables, we indicated the terms that characterized each tissue, i.e. the GO terms identified as ‘over-correlated’ in the same tissue for all the comparisons. As an example, a term is related to heart if it is ‘over-correlated’ in heart in both comparisons: heart versus skeletal and heart versus smooth. We included also a ‘striated’ label for the terms common to heart and skeletal muscles, the ‘striated’ tissues.
As far as the cellular component is concerned, we wished to identify any possible heart-specific component of ‘extracellular matrix’ (GO:0031012). The results from Fun&Co are shown in Table 2 and the role of extracellular matrix seems obvious: laminin and collagens are strongly represented in skeletal muscle, while the association of pre- and post-synaptic membranes component in the cardiac cells likely relates to their conductivity.
|
In the second GO category, biological processes, we investigated the signaling properties of heart, skeletal and smooth muscles (see Table 3 in Supplementary Material). Considering ‘cell surface receptor linked signal transduction’ (GO:0007166) as GOmain, we noticed that transmembrane receptor protein tyrosine phosphatase and dopamine receptors appeared associated to skeletal muscle, while IGF1R is associated to heart.
Transgenic mice over-expressing IGF1R (Insuline growth factor like 1 receptor) in the heart displayed cardiac hypertrophy which was due to an increase in myocyte size, and there was no evidence of histopathology. This study suggests that targeting the cardiac IGF1R-PI3K(p110alpha) pathway could be a potential therapeutic strategy for the treatment of heart failure (Canicio and Kaliman, 2001). For insulin receptor signaling pathway, we obtained a P-value of 0.303 in the comparison between heart and skeletal muscle and a P-value of 0.207 in the comparison between heart and smooth muscle. Performing a similar experiment with GlobalTest, the comparative P-values were respectively, 0.508 and 0.226.
In skeletal muscle, we noticed the presence of transmembrane receptor protein tyrosine phosphatase signaling pathway seems to be obvious: protein-tyrosine phosphatases (PTPases) have an important role in the regulation of insulin signal transduction, and the skeletal muscle is the major site of tissue insulin resistance in obesity and diabetes. The PTPase activity in skeletal muscle from non-diabetic obese subjects was increased significantly by 40–70% compared to the level in controls (Ahmad et al., 1997). We obtained a P-value of 0.149 in the comparison between heart and skeletal muscle and a P-value of 0.379 in the comparison between heart and smooth muscle. Performing a similar experiment with GlobalTest, the comparative P-values were respectively, 0.297 and 0.918.
In ‘intracellular signaling cascade’ (GO:007242), the activation of NF-
B-inducing kinase pathway appears to be skeletal muscle associated. We obtained a P-value of 0.279 in the comparison between cardiac and skeletal muscle. Performing a similar experiment with GlobalTest, the comparative p was 0.748.
The third and last GO category analyzed, molecular functions, yielded many additional interesting results (see Table 4 in Supplementary Material). The first general GO term analyzed in this category was ‘kinase activity’ (GO:0016301). The ephrin receptor activity was the only one specific to cardiomyocytes: this protein is involved in cell–cell communication during development and in particular EphA3 plays a critical role in heart development (Stephen et al., 2007).
In striated muscle, which includes both heart and skeletal muscle, the IGF receptor activity is listed: this receptor regulates the cell growth and development in muscles and other tissues. In heart, IGFs are locally produced and modulates cardiomyocyte growth and maturation. Biochemical alteration (expression variation of IGFs) may be associated to fetal/neonatal growth abnormalities of rats (Engelmann et al., 1989). We obtained a P-value of 0.561 in the comparison between skeletal and smooth muscle and a P-value of 0.524 in the comparison between heart and smooth muscle. Performing a similar experiment with GlobalTest, the comparative P-values were respectively, 0.699 and 0.754.
In smooth muscle on the other hand, MAP kinase activity was apparent in its multiple steps. MAP kinase is part of a signal transduction pathway that promotes cell divisions in response to extracellular stimuli. MAP kinase pathway, activated by angiotensin II, is involved in hypertensive vascular remodeling, associated with cell growth and increased deposition of extracellular matrix, in particular collagen (Touyz et al., 2001). Furthermore, the G protein coupled receptors kinase activity, also short-listed in smooth muscle cells, mediates, via the MAPK pathways, the mitogenic effects of oxidized low-density lipoprotein on vascular smooth muscle cells (Yang et al., 2001). This mechanism is proven to be involved in pathogenesis of atherosclerosis.
Other kinase activities could be particularly related to the different metabolic pathways of these three different muscle tissues. In skeletal muscles, the phosphorylase kinase activity is associated with the glycogen metabolism. Glycogen representing an easily available source of glucose. The liver and the skeletal muscles are in fact the main tissues that stock the glycogen, and the glycogen phosphorylase kinase is the key regulatory enzyme in this process.
In striated, a range of kinase activities related to glucose metabolism (glycolisis and gluconeogenesis) were apparent, e.g. hexokinase and other enzymes, like pantothenate kinase, involved in synthetic pathways of acetyl-CoA, the point of connection of the main metabolic oxidative pathways (amino acids, fatty acids and carbohydrates).
In stimulated smooth cells, the concentration of diacylglycerol (DAG) rises rapidly, and DAG functions as a second messenger by activating protein kinase C, which in turn regulates many cellular responses, including growth and differentiation. The attenuation of the DAG signal and phospholipid synthesis, by the conversion of DAG to phosphatidic acid (PA), is regulated by DAG kinases (DGKs). PA can also serve as a lipid messenger and the net effect of conversion of DAG to PA might vary from cell to cell and condition to condition.
Diacylglycerol kinase (DGK) phosphorylates the lipid second messenger DAG to phosphatidic acid. DGK-theta is present both in smooth muscle and in endothelial cells of the small blood vessels. DGK-theta activity can be increased by noradrenaline (NA) and this pathway is thought to have a physiological role in vascular smooth-muscle responses (Walker et al., 2001).
Guanylate kinase catalyzes the phosphorylation of either GMP to GDP or dGMP to dGDP and is an important enzyme in nucleotide metabolic pathways. Co-expression of guanylate kinase with thymidine kinase enhances pro-drug cell killing in vitro and suppresses vascular smooth muscle cell proliferation in vivo (Akyurek et al., 2001).
In striated, the inositol trisphosphate 3-kinase converts Ins-1,4,5-P3 to Ins-1,3,4,5-P4, that modulates the entry of Ca2+ from an extracellular source. The 3-kinase activity is significantly activated by the Ca2+/calmodulin complex. In some experiments, the IP3 kinase activity was increased in SHRSP (stroke-prone spontaneously hypertensive rats) and its activity was markedly affected by divalent cations. These data suggest that the accumulations of IP3 and IP4 after hormonal stimulation play a physiologic role, possibly by alteration of Ca2+ levels in cardiac tissue (Kawaguchi et al., 1990).
Another GO term that provided interesting results was ‘G-Protein coupled receptor activity‘ (GO:0004930). In heart, the thrombin activity induces IP3 formation associated to increase in cytosolic calcium, enhanced automaticity and prolong repolarization: this can be related to the electrical abnormalities observed in ischemia and infarction (Steinberg et al., 1991). In skeletal muscle the prostaglandin activity is important: arachidonic acid metabolites, such as prostaglandins (PG), are regulatory of vascular tone and can be released from the contracting muscles under the influence of dynamic exercise (Karamousi et al., 2001). This can in turn augment the blood flow and allow the incoming of nutrients and oxygen, to support an increasing metabolic muscle request. Furthermore, prostaglandin F2 is involved in the multi-step process leading to the formation of large multinucleated muscle cells. Therefore, the use of prostaglandins might be therapeutic for treatment of muscle loss due to aging, injury and disease. And conversely caution should be taken in using drugs that inhibit PG production (like e.g. non-steroidal anti inflammatory drugs) which may be deleterious for muscle growth (Horsley and Pavlath, 2003). Also in smooth prostaglandins are active, but in this case prostaglandin E receptor activity induces relaxation, e.g. in trachea Platelet activating factor (PAF) receptor activity is another term specific to skeletal muscle: PAF cumulative effects in skeletal muscle reduce protein synthesis during endo-toxic and septic shock (Karlstad et al., 2000). In addition, PAF seems to be involved in skeletal muscle ischemia-reperfusion injury (IRI): infusion of PAF antagonists into the muscle prior to reperfusion can indeed reduce muscle necrosis (Silver et al., 1996).
On the basis of opioid-stimulated contraction of dispersed gastric smooth muscle cells, it has been suggested that these cells possess opioid receptors of three subtypes: kappa, mu and delta. In smooth cells, the disorder of Ca2+ regulation induced by hemorrhagic shock was mediated by opioid receptor and alpha-adrenoceptor, which may be partly responsible for the vascular hyporesponse, and opioid receptor antagonists improved the response of resistance arteries to vascular stimulants in decompensatory stage of hemorrhagic shock (Kai et al., 2004).
Vaso-active intestinal polypeptide receptor is involved in smooth muscle relaxation and, in particular, in bladder, stomach and the esophageal sphincter. Tachykinin receptor activity in striated is due to its role: takykinin has cardio-acceleratory effect (Silwowska et al., 2001) and probably some coordinated effect on skeletal muscle.
Again in striated, notable is the growth hormone (GH)- releasing hormone activity. Treatment with GH enhances intrinsic cardiac myocyte contractile function, restores myocardial sarcoplasmic reticulum Ca2-ATPase expression levels and increases myocardial capillary density in the failing heart, promoting an adaptive form of cardiac hypertrophy. In addition, it promotes favorable non-cardiac effects in chronic hearth failure, reducing skeletal muscle atrophy, enhancing skeletal muscle strength and correcting vascular and endothelial dysfunction (Wollert and Drexler, 2003). We obtained a P-value of 0.529 in the comparison between heart and smooth and a P-value of 0.449 in the comparison between skeletal muscle and smooth. Performing a similar experiment with GlobalTest, the comparative P-values were respectively, 0.78 and 0.861.
|
5 CONCLUSION |
|---|
|
TOPABSTRACT1 INTRODUCTION2 APPROACH3 METHODS4 RESULTS AND DISCUSSION5 CONCLUSIONACKNOWLEDGEMENTSREFERENCES |
|---|
Fun&Co is a novel and very efficient way of mining functional differences from a number of datasets in a pairwise manner. The application extracts the most significant differences from the molecular expression data, as shown in this article on skeletal, heart and smooth muscles. The results are highly informative and synthetic. Figure 4 highlights the components of the heart signaling network (Heineke and Moalkentin, 2006) which were identified by Fun&Co. It is apparent that as many as a dozen critical points were correctly detected. This finding supports the potential usefulness of this application in the high level analysis of transcriptome. Fun&Co is implemented as a web service and publicly available.
|
|
ACKNOWLEDGEMENTS |
|---|
|
TOPABSTRACT1 INTRODUCTION2 APPROACH3 METHODS4 RESULTS AND DISCUSSION5 CONCLUSIONACKNOWLEDGEMENTSREFERENCES |
|---|
Telethon and MIUR (PRIN) for support. GebbaLab is a PRRIITT project by Regione Emilia-Romagna.
Conflict of Interest: none declared.
|
FOOTNOTES |
|---|
Associate Editor: John Quackenbush
1See: http://www.geneontology.org/ 
2See: http://www.ncbi.nlm.nih.gov/geo/
3See http://froi.sourceforge.net/documents/technical/matlab/FDR.html
4See: http://www.affymetrix.com/analysis/index.affx
5For more details on the comparative p evaluation, see the vignette of GlobalTest R-package.
Received on June 29, 2007; revised on August 13, 2007; accepted on August 14, 2007
|
REFERENCES |
|---|
|
TOPABSTRACT1 INTRODUCTION2 APPROACH3 METHODS4 RESULTS AND DISCUSSION5 CONCLUSIONACKNOWLEDGEMENTSREFERENCES |
|---|
Ahmad F, et al. Alterations in skeletal muscle protein-tyrosine phosphatase activity and expression in insulin-resistant human obesity and diabetes. J. Clin. Invest (1997) 100:449–458.[Web of Science][Medline]
Akyurek LM, et al. Coexpression of guanylate kinase with thymidine kinase enhances prodrug cell killing in vitro and suppresses vascular smooth muscle cell proliferation in vivo. Mol. Ther (2001) 3:779–786.[CrossRef][Web of Science][Medline]
Al-Shahrour F, et al. FatiGO: a web tool for finding significant associations of Gene Ontology terms with groups of genes. Bioinformatics (2004) 20:578–580.
Al-Shahrour F, et al. Discovering molecular functions significantly related to phenotypes by combining gene expression data and biological information. Bioinformatics (2005) 21:2988–2993.
Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J. Stat. R. Soc (1995) 57:289–300.
Canicio J, Kaliman P. Nuclear factor kappa B-inducing kinase and Ikappa B kinase-alpha signal skeletal muscle cell differentiation. J. Biol. Chem (2001) 276:20228–20233.
Draghici S, et al. Global functional profiling of gene expression. Genomics (2003) 81:98–104.[CrossRef][Web of Science][Medline]
Eisen MB, et al. Cluster analysis and display of genome-wide expression patterns. Proc. Natl Acad. Sci. USA (1999) 95:14863–14868.[CrossRef][Web of Science]
Engelmann GL, et al. Insulin-like growth factors and neonatal cardiomyocyte development: ventricular gene expression and membrane receptor variations in normotensive and hypertensive rats. Mol. Cell Endocrinol (1989) 63:1–14.[CrossRef][Web of Science][Medline]
Goeman JJ, van Houwelingen HC. A global test for groups of genes: testing association with a clinical outcome. Bioinformatics (2004) 20:93–99.
Heineke J, Molkentin JD. Regulation of cardiac hypertrophy by intracellular signalling pathways. Nat. Rev. Mol. Cell Biol (2006) 8:589–600.
Horsley V, Pavlath GK. Prostaglandin F2(alpha) stimulates growth of skeletal muscle cells via an NFATC2-dependent pathway. J. Cell Biol (2003) 161:111–118.
Hosack DA, et al. Identifying biological themes within lists of genes with EASE. Genome Biol (2003) 4:R70.[CrossRef][Medline]
Kai L, et al. Opioid receptor antagonists increase [Ca2+]i in rat arterial smooth muscle cells in hemorrhagic shock. Acta Pharmacol. Sin (2004) 25:395–400.[Web of Science][Medline]
Karamouzis M, et al. In situ microdialysis of intramuscular prostaglandin and thromboxane in contracting skeletal muscle in humans. Acta Physiol. Scand (2001) 171:71–6.[CrossRef][Web of Science][Medline]
Karlstad MD, et al. Platelet-activating factor (PAF)-induced decreases in whole-body and skeletal muscle protein synthesis. Shock (2000) 14:490–498.[Web of Science][Medline]
Kawaguchi H, et al. Inositol trisphosphate kinase activity in hypertrophied rat heart. Biochem. Med. Metab. Biol (1990) 44:42–50.[CrossRef][Web of Science][Medline]
Lee HK, et al. ErmineJ: tool for functional analysis of gene expression data sets. BMC Bioinformatics (2005) 6:269.[CrossRef][Medline]
Mootha VK, et al. PGC-1alpha-responsive genes involved in oxidative phosphorylation are coordinately downregulated in human diabetes. Nat. Genet (2003) 34:267–273.[CrossRef][Web of Science][Medline]
Pavlidis P, et al. Exploring gene expression data with class scores. Pac. Symp. Biocomput (2002) 474–485.
Rosner B. Fundamentals of Biostatistics (2000) Pacific Grove, CA: Duxbury.
Silver D, et al. Role of platelet-activating factor in skeletal muscle ischemia-reperfusion injury. Adv. Exp. Med. Biol (1996) 416:217–221.[Medline]
Sliwowska J, et al. Cardioacceleratory action of tachykinin-related neuropeptides and proctolin in two coleopteran insect species. Peptides (2001) 22:209–217.[CrossRef][Web of Science][Medline]
Steinberg SF, et al. Thrombin modulates phosphoinositide metabolism, cytosolic calcium, and impulse initiation in the heart. Circ. Res (1991) 68:1216–1229.
Stephen LJ, et al. A critical role for the EphA3 receptor tyrosine kinase in heart development. Dev. Biol (2007) 302:66–79.[CrossRef][Web of Science][Medline]
Touyz RM, et al. p38 Map kinase regulates vascular smooth muscle cell collagen synthesis by angiotensin II in SHR but not in WKY. Hypertension (2001) 37:574–580.
Volinia S, et al. GOAL: automated Gene Ontology analysis of expression profiles. Nucleic Acids Res (2004) 32:W492–W499.
Walker AJ, et al. Diacylglycerol kinase theta is translocated and phospho-inositide 3-kinase-dependently activated by noradrenaline but not angiotensin II in intact small arteries. Biochem. J (2001) 353:129–137.[CrossRef][Web of Science][Medline]
Wollert KC, Drexler H. Growth hormone and proinflammatory cytokine activation in heart failure. Just a new verse to an old sirens' song? Eur. Heart J (2003) 24:2164–2165.
Yang CM, et al. Mitogenic effect of oxidized low-density lipoprotein on vascular smooth muscle cells mediated by activation of Ras/Raf/MEK/MAPK pathway. Br. J. Pharmacol (2001) 132:1531–1541.[CrossRef][Web of Science][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||